Review moderation on a website

Review moderation on a website is a pre-publication queue, or fast post-publication handling, plus a complaint tool. Without it, TrustScore breaks: spam, insults, paid negativity, and batches of five-star reviews. A review service script provides statuses and roles; the rules and moderator discipline are your responsibility.

Related pages: review service, website reviews, company rating, widget, photos.

Premoderation or postmoderation

Premoderation: the review becomes public only after approval. It is calmer at launch and in toxic niches. Postmoderation: it appears immediately, and complaints plus anti-spam clean it later. This is faster for authors, riskier for platform reputation. A hybrid works well: new accounts go through premoderation, trusted accounts through postmoderation.

Publish the response time. An SLA of "up to 24 hours" is better than proud "instant" moderation you cannot maintain.

Moderator queue

The list should show date, company, rating, text fragment, anti-spam flags, and complaint count. The card opens in one click. Actions: approve, reject with a reason, request an edit, block the author. Standardize rejection reasons so the author receives a clear message, not just "rejected".

• Insults and hate speech.
• Personal data of third parties.
• Advertising and irrelevant links.
• Clear manipulation or copy-paste.
• Review about the wrong company.
• Threats and legally prohibited content.

Complaints from brands and users

Add a "report" button next to each review. Reasons differ for a brand and for a reader. A complaint must not automatically delete the text. The moderator sees complaint history. Repeated false brand complaints on every negative review are a reason to discuss the rules, not to remove reviews in bulk.

After hiding a review, recalculate the score and invalidate the widget. Notify the author. Notify the brand too if it filed the complaint.

Representative reply

Brand replies can be premoderated or published immediately with postmoderation, depending on your risk tolerance. An offensive representative reply is also hidden: double standards destroy trust in the platform. Representative verification is required before the first reply.

Anti-spam

Rate limits, captcha for new accounts, stop words, link checks, and duplicate paragraph detection. Automation is an assistant, not a judge: a human makes the final decision in disputed cases. Review auto-rejection logs so you do not kill honest reviews written in formal language.

Moderate photos separately: prohibited images, other people's passports, screenshots with phone numbers. See photo reviews.

Roles and conflicts of interest

A moderator should not also be a company representative in the same system without strict restrictions. The admin sees an audit log: who approved, who hid, and when. Without a log, a scandal six months later is impossible to investigate.

Rules and training

A one-page policy with approve/reject examples is more useful than a thick PDF. Review disputed cases with the team once a month. Written precedents reduce chaos when several moderators work shifts.

Acceptance checklist

1. A new review enters the queue, not the public showcase, when premoderation is enabled.
2. Approval changes the profile score and widget.
3. Rejection with a reason is sent to the author.
4. A complaint creates a task; hiding recalculates the rating.
5. The brand reply appears in the card according to your rules.
6. The audit log records moderator actions.

Do not promise companies "complete removal of any negativity". Promise fair moderation by rules. That is the only position a platform can hold for a long time.

Shifts, load, and queue priorities

Do not sort the queue only FIFO. Priority: complaints with many flags, reviews about companies in sensitive niches, brand replies with stop words, and new accounts. Ordinary premoderation of calm texts can be handled in a second pass. Otherwise moderators drown in small items while a scandal stays public.

Use "reviews per hour" targets carefully: they encourage mechanical approval. Better metrics are sample-check quality and the number of justified appeals. An author's appeal against rejection should be a separate short queue, not the same stream.

Night automatic rules for stop words and links are acceptable, but morning human review of false positives is mandatory. Log exactly what triggered, otherwise automation becomes a myth that "does something by itself".

Transparency for authors and brands

Statuses in the author's account: under review, published, rejected, hidden by complaint. Brands see new reviews and outcomes of their complaints. Silent disappearance of text without a status encourages conspiracy theories and reposting of the same toxic copy-paste.

Once a quarter, publish anonymized statistics: how many reviews were accepted, how many rejected, and median moderation time. This is a strong trust signal for press and partners if you keep the numbers honest.

Training new moderators on a golden set of 50 real cases with reference decisions is faster than any presentation. Update the set when rules change.

Email integration and escalation

Email alerts to moderators about an overloaded queue, with threshold N, save weekends. Escalation to a senior moderator for complaints such as "defamation / personal data" should be a separate button, not a request to write to someone in a messenger. Record the response deadline in the escalation card.

Check author email templates for tone: dry, specific, and without moralizing. A template that says "your review was rejected because you are bad" triggers repeat spam. A concrete reason and a link to rules reduce repeats.

Once per sprint, export 20 random moderation decisions for review. Catch rule drift between shifts before a public scandal does. Feed the review results back into the golden case set.

Photo connection: if the text is fine but the image is not, use partial hiding. Rejecting a useful review completely because of one extra screenshot is a bad habit that drives authors away.

Boundary with support and legal

A moderator hides content by platform rules. Legal joins when an official claim or personal data request arrives. Support answers "why was this rejected" without rewriting the policy on the fly. Three roles, three channels. Mixing them in one chat turns yesterday's exception into today's "but you allowed it for someone else". Record exceptions in writing and with an expiration date if you need them at all.

Keep an instruction for what to do if a review reaches the media. The platform's response speed in a public conflict matters as much as queue speed on a calm day.